In recent years, university, government, business, and financial service entities, among others, have increasingly relied upon data center networks that incorporate racks of server computers (“servers”) to implement application programs (“applications”) for supporting their specific operational requirements, including, but not limited to, data base management applications, document and file sharing applications, searching applications, gaming applications, and financial trading applications. Such data center networks are generally expanding in terms of the number of servers incorporated therein, as well as the networking equipment needed to interconnect the servers for accommodating the data transfer requirements of the respective applications.
Conventional data center networks typically have hierarchical architectures, in which each server co-located in a particular rack is connected via one or more Ethernet connections to a top-of-rack Ethernet switch (the “top-of-rack switch”). A plurality of such top-of-rack switches form what is referred to herein as the “access layer”, which is generally the lowest level of the hierarchical network architecture. The next higher level of the hierarchy is referred to herein as the “aggregation layer”, which can include a plurality of Ethernet switches (the “aggregation switch(es)”) and/or Internet protocol (IP) routers. Each top-of-rack switch in the access layer can be connected to one or more aggregation switches and/or IP routers in the aggregation layer. The highest level of the hierarchy is referred to herein as the “core layer”, which generally includes a plurality of IP routers (the “core switches”) that can be configured to provide ingress/egress points for the data center network. Each aggregation switch and/or IP router in the aggregation layer can be connected to one or more core switches in the core layer, which, in turn, can be interconnected to one another. In such conventional data center networks, the interconnections between the racks of servers, the top-of-rack switches in the access layer, the aggregation switches/IP routers in the aggregation layer, and the core switches in the core layer, are typically implemented using point-to-point Ethernet links.
Although conventional data center networks like those described above have been employed to satisfy the operational requirements of many university, government, business, and financial service entities, such conventional data center networks have drawbacks. For example, data communications between servers that are not co-located within the same rack may experience excessive delay (also referred to herein as “latency”) within the data center networks, due to the multitude of switches and/or routers that the data may be required to traverse as it propagates “up”, “down”, and/or “across” the hierarchical architecture of the networks. Data communications between such servers may also experience latency within the respective switches and/or routers of the data center networks due to excessive node and/or link utilization. Further, because multiple paths may be employed to deliver broadcast and/or multicast data to different destinations within the data center networks, such broadcast and/or multicast data may experience excessive latency skew. Such latency and/or latency skew may be exacerbated as the sizes of the data center networks and/or their loads increase.
In addition, conventional data center networks typically include network management systems that employ configuration data for proper allocation of computing resources within the data center networks. However, such configuration data frequently lack contextual information, such as how the topology of a data center network should be configured in view of the available computing resources to achieve a desired level of application performance. For example, such network management systems may employ the Open Virtualization Format (also referred to herein as the “OVF standard”) to facilitate the control and provisioning of such computing resources. However, the OVF standard generally lacks contextual information pertaining to the network topology, and may therefore be incapable of assuring that the available computing resources are being properly provisioned for the desired application performance level. As a result, problems with latency, data bottlenecks, etc., may be further exacerbated, thereby slowing down or otherwise inhibiting data movement within the data center networks.
It would therefore be desirable to have data center network architectures that avoid at least some of the drawbacks of the conventional data center networks described above.